Make a 3D game with AI, and know exactly which part the AI did

The advertised version of AI game making is that you type a paragraph and receive a game. What you actually receive from that approach is a scene: some terrain, some props, a character that moves. It is a demo, and the second you want to change something specific about it you are back to describing shapes in English, which is a bad way to describe shapes.

The honest version is narrower and more useful. AI writes code well. Code is text, it has a right answer, and a model that has read a great deal of GDScript can produce a working interaction loop faster than a person learning the language can produce a broken one.

So Director 1 writes the code. Everything about how the world looks and where things are is yours, because that is the part where a sentence is a worse tool than a mouse.

You describe behaviour, not geometry. "When the player steps into the ring of stones, the fog closes in and the door on the hill unlocks." Director 1 writes GDScript in your project that does that.

Then you play it, and the fog is too fast, or the door unlocks silently and you never notice it happened. You say that. It gets changed. Over an evening this converges on something that plays the way you imagined, and the reason it converges is that you are supplying the judgement at every step.

You never have to read the GDScript. It is there, in your project, in normal files, if you or anyone else wants to.

Terrain gets sculpted. Water takes an outline you draw. Weather and sky are properties of the world you set. Buildings, props and characters get placed where you put them.

This is deliberate and it is the opposite of most AI game tools, which try to generate the world and give you a text box to complain about it. Placing a tree is a two-second operation with a mouse and a frustrating one with a prompt. Nothing is gained by routing it through language.

Assets themselves are a different matter — those come from generation, from the web, from your own files, or get built directly as geometry when that is the quickest honest route to something you can play.

Stage Engine is a desktop application for Mac and Windows. Godot ships inside it, and your game runs on your own hardware when you press play. No streaming, no queue, no browser.

Building needs the internet, because Director 1, asset generation and world data are all network services. This is not an offline tool and will not pretend to be one. The line worth remembering: building is online, playing is not.

The app is free to download and there is a free plan you can build on. Indie is $10 a month, Pro is $50. There is no team or enterprise tier — this is a desktop app for one person.

The paid plans buy more AI usage. They do not unlock a better engine, a bigger world size, or an export option. Everyone is on the same Godot.

Movement, before anything else. You cannot judge a world you cannot walk through, and every later decision about scale depends on how fast the character moves. Get it right while there is nothing else to distract you.

Then one interaction. Something you can press, pick up, or open. This proves the loop between describing behaviour and playing it works, and it is where you learn how specific you need to be.

Then whatever your game is actually about. By this point you know how the tool responds and you can describe things in bigger pieces, which is both faster and cheaper than a hundred small adjustments.